Posterior analysis of random taste coefficients in air travel behaviour modelling

Increasing use is being made of random coefficients structures, such as mixed logit, in the analysis of air travel choice behaviour. These models have the advantage of being able to retrieve random variations in sensitivities across travellers. An important issue, however, arises in the computation of willingness to pay indicators, such as the valuation of travel time savings, on the basis of randomly distributed coefficients. Indeed, with the standard approach of using simulation of the ratios across random draws, major problems can be caused by outliers, leading to biased trade-offs, which in turn lead to major issues in policy analyses. Here, a different approach is explored, making use of individual-specific draws from the random distributions, conditioned on the observed sequence of choices for each respondent. An analysis making use of stated preference data for airport and airline choice confirms the advantages of the approach using conditional draws, producing much more realistic distributional patterns for a range of willingness to pay indicators.